Speaker identification embeddings audio fragment length I have a base of audio samples matched with concrete speaker like
